ChatGPT Atlas
Today we’re introducing ChatGPT Atlas, a new web browser built with ChatGPT at its core.
[…]
With Atlas, ChatGPT can come with you anywhere across the web—helping you in the window right where you are, understanding what you’re trying to do, and completing tasks for you, all without copying and pasting or leaving the page. Your ChatGPT memory is built in, so conversations can draw on past chats and details to help you get new things done.
[…]
ChatGPT Atlas is launching worldwide on macOS today to Free, Plus, Pro, and Go users. Atlas is also available in beta for Business, and if enabled by their plan administrator, for Enterprise and Edu users. Experiences for Windows, iOS, and Android are coming soon.
Alas, it doesn’t support AppleScript and has System Settings–style preferences.
Atlas, like Perplexity’s Comet, is a Chromium-based browser. You cannot use it without signing in to ChatGPT.
[…]
The company says it only retains pages until they have been summarized, and I am sure it thinks it is taking privacy as seriously as it can. But what about down the road? What could it do with all of this data it does retain — information that is tied to your ChatGPT account?
The new tab page is predictably a text box that intelligently does what you ask it to do, routing your queries to perform web searches, start a standard ChatGPT chat, or simply load a website from your bookmarks or history. You can, of course, also just paste in the URL and go.
[…]
I’m also a big proponent of the “show full URL in address bar” feature in all browsers, and I’m happy to see this is here as well. It’s a little thing, but I’m always worried it’s on its way out.
[…]
The app does not have an agent mode as of yet, but it sounds like that will be coming in the relatively near future. My experience with these modes in other browsers has been a major letdown, so we’ll see if OpenAI can do any better, but I’m not holding my breath here.
I use the ChatGPT app at work, and I actually like having a separate window for all A.I. shenanigans: I can switch apps quickly, I can close it, and I can call it with a keyboard shortcut. Sure, it’s way more limited, and I need to jump from one app to another more often, but I actually see this as a feature.
This is not just about Atlas; I haven’t read about any cool use case of an A.I. browser, whether it is Dia or Comet. Maybe this new browser will change things, maybe it will reach more people and we will see good examples, but so far, it feels like even folks at OpenAI struggled to find compelling use cases. Or maybe I was too bored by the video to pay attention?
Previously:
9 Comments RSS · Twitter · Mastodon
Hello — I’m John and am on the Atlas team (and am devoted mjtsai reader!). Would love to hear about what use cases with AppleScript you’d like to see. Being more developer friendly is definitely on our radar.
Feel free to get in touch on Twitter (dms are open)
Small note about Matt Birchler’s quote — there is an agent mode already available. Not available for free plans at the moment though.
Hello Jon Nastros, I'm Dave.
(1) Are you a real human being? I am. Glad to hear you are a devoted reader of this blog. It's really hard anymore to discern this sort of thing online, and using a browser that is "AI" based - and requires a login (!?) kind of sounds.... off.
(2) Glad to hear "you" - real or AI - wits to be more developer friendly. How about telling me more? Maybe here - and not Twitter (glad you didn't say X). Better yet - maybe just simply DO IT. Or it that beyond you? Why should I give you "use cases" for anything? Particularly over "Twitter" - which ceased to exist a few years back.
(3) Small note about.... whatever.... what is this "agent mode already available"? And why isn't a "free plan" available at the moment?
I see John here so I don't want to be too harsh--thank you John for your work on Atlas. I know there's a lot of great people doing great work at OpenAI. ChatGPT is indeed my go-to AI, so I'm not a hater.
I want to bring up though that I believe this to be part of OpenAI's broader push to harvest data directly from users' real world experiences to continue improving their models. It's pretty clear that they are plateauing in performance due to lack of high quality data, and the best way to get more data that is reliably human is to have harvesting devices of some kind, like browsers or handheld devices (e.g., https://builtin.com/articles/openai-device). I think that's why they're partnering with Jony Ive and I'm guessing that's why they're suddenly interested in having a browser.
They may make commitments to safety or privacy for the browser now, but they have repeatedly reneged on their pledges. Remember when they were non-profit only, before Sam schemed to build a for-profit segment (see e.g., https://clsbluesky.law.columbia.edu/2024/03/05/the-untold-nonprofit-story-of-openai/)? Remember when safety was supposed to be their #1 priority, before they told GPT to favor engagement instead of safety (see e.g., https://www.theguardian.com/technology/2025/oct/22/openai-chatgpt-lawsuit)? In my opinion, they have a lot of work to do before people should trust increasing OpenAI's access via a browser or similar. They might earnestly be focusing on privacy now, but how can anyone be assured that once the profit motivation becomes large enough they won't pivot and begin extracting what value they can from their browser? They're losing $10 billion a quarter or something like that; that possible revenue (whether via selling data gained via the browser, access to users via ads in GPT/browser, or acquiring more human data to continue to improve GPT) is going to start to look really tempting.
I tried Dia and when I realized the o my thing it could do is search and summarize websites I was disappointed and baffled. Who cares?
That’s not a very useful feature, certainly not something worth building a whole browser around
Agent automation is the only useful thing I can think of
First thing I noticed that makes it a complete nonstarter for me:
- No support for iCloud passkeys. Most browsers (Chrome, Edge, Firefox, Brave, Vivaldi, etc) support these out of the box, no extension required.
- iCloud Passwords extension doesn’t work, just says it can’t communicate with a helper application. Again, this works great with all the browsers I mentioned above.
I don’t get how you can release a macOS-only web browser that doesn’t work with any of Apple’s credential ecosystem. Even if the AI features are great, this makes it unusable.
Maybe I’m worried about nothing, but it’s a hard leap for me to give ChatGPT access to all my logged in accounts. I’m still uncomfortable with it having plain text access to everything. I use ChatGPT all the time but there’s still something creepy about it having logins to all my accounts. It seems like a useful research tool but I still wouldn’t log in to anything sensitive with it at this point.
Am I crazy?
I'm more interested in Googles idea to put a small LLM in the browser that devs can use to do small things on the client side rather than having to pay for tokens.
I've no idea what to do with an ai browser.
Getting older by the day
